Electromyography as a Recording System for Eyeblink Conditioning with Functional Magnetic Resonance Imaging

NeuroImage 17, 977–987 (2002)doi:10.1006/nimg.2002.1199

Electromyography as a Recording System for Eyeblink Conditioningwith Functional Magnetic Resonance Imaging

M-G. Knuttinen,* T. B. Parrish,† C. Weiss,* K. S. LaBar,‡ D. R. Gitelman,†,§ J. M. Power,*M-M. Mesulam,§ and J. F. Disterhoft*,1

Northwestern Cognitive Brain Mapping Group, *Department of Physiology, †Department of Radiology, and §Department of Neurology,Northwestern University Medical School, Chicago, Illinois 60611; and ‡Center for Cognitive Neuroscience,

Duke University, Durham, North Carolina 27708

Received March 29, 2001

This study was designed to develop a suitablemethod of recording eyeblink responses while con-ducting functional magnetic resonance imaging(fMRI). Given the complexity of this behavioral setupoutside of the magnet, this study sought to adapt andfurther optimize an approach to eyeblink condition-ing that would be suitable for conducting event-re-lated fMRI experiments. This method involved the ac-quisition of electromyographic (EMG) signals from theorbicularis oculi of the right eye, which were subse-quently amplified and converted into an optical signaloutside of the head coil. This optical signal was con-verted back into an electrical signal once outside themagnet room. Electromyography (EMG)-detected eye-blinks were used to measure responses in a delayeyeblink conditioning paradigm. Our results indicatethat: (1) electromyography is a sensitive method forthe detection of eyeblinks during fMRI; (2) minimalinteractions or artifacts of the EMG signal were cre-ated from the magnetic resonance pulse sequence; and(3) no electromyography-related artifacts were de-tected in the magnetic resonance images. Further-more, an analysis of the functional data showed areasof activation that have previously been shown inpositron emission tomography studies of human eye-blink conditioning. Our results support the strength ofthis behavioral setup as a suitable method to be usedin association with fMRI. © 2002 Elsevier Science (USA)

INTRODUCTION

The eyeblink conditioning paradigm is a particularlywell-understood model system to study the mecha-nisms of associative learning and memory in animals(Disterhoft et al., 1977; Thompson et al., 1976) and in

1 To whom correspondence should be addressed at Department ofPhysiology, Northwestern University Medical School, 303 East Chi-cago Avenue, Searle 4-427, Chicago, IL 60611.

977

humans (Carrillo et al., 1997; Daum et al., 1993; Daumand Schugens, 1996; Gabrieli et al., 1995; McGlinchey-Berroth et al., 1997; Woodruff-Pak, 1988, 1993). Eye-blink conditioning requires that the subject associate aneutral stimulus, e.g., auditory conditioned stimulus(CS), with a behaviorally significant stimulus, e.g., acorneal airpuff unconditioned stimulus (US). The bestunderstood eyeblink conditioning paradigm is the de-lay paradigm, which involves associating a tone CSthat precedes, overlaps, and coterminates with a cor-neal airpuff US. Eyeblink conditioning in humans re-quires that eyeblink responses be monitored as thesubject responds to the stimuli. The majority of thestudies involving human eyeblink conditioning haveused reflectance of infrared light as the method ofchoice to measure eyelid responses (Carillo et al., 1997;Clark and Squire, 1998). However, since the electroniccomponents involved in the infrared measurement in-terfere with imaging in the MRI environment (Thomp-son et al., 1994), we have developed a method to detectthe electromyographic (EMG) activity of the eyelids(orbicularis oculi). In a study reported elsewhere, it hasbeen shown outside of the magnet that the two tech-niques are equivalent in their ability to detect eye-blinks (Knuttinen et al., 2001). The purpose of thisstudy was to optimize the EMG approach as a suitablemethod for detecting eyeblink responses during fMRI,and to further develop a behavioral setup that can beused to study eyeblink conditioning in the magnet.

Functional MRI provides detailed images that reflectchanges in cerebral blood flow, cerebral blood volume,and oxygenation induced by sensory, motor, or cogni-tive tasks (Cohen and Bookheimer, 1994; Kwong, 1995;LeBihan and Karni, 1995), including eyeblink condi-tioning (Knuttinen et al., 2000; Preston et al., 2000;Ramnani et al., 2000). This imaging approach hasserved as a valuable tool in detecting patterns of he-modynamic changes in frontal cortex, cerebellum, andhippocampus and associated temporal lobe structureswhile subjects are performing a variety of learning

1053-8119/02 $35.00© 2002 Elsevier Science (USA)

All rights reserved.

tasks (Demb et al., 1995; Desmond et al., 1997; Gabrieliet al., 1996, 1997; Poldrack et al., 1998; Wagner et al.,1997). Eyeblink conditioning is well-suited for neuro-biological analyses with fMRI due to the use of appro-priate unpaired control (pseudoconditioning) proce-dures that can distinguish between areas of the brainthat are responding to sensory stimuli (e.g., auditoryand somatosensory cortices) and those that are selec-tively activated under paired learning conditions bycomparing pseudoconditioning and conditioning trialsin the same subject. Subjects can also return for retest-ing over a period to examine the processes underlyingmemory consolidation. The dynamic characterizationsof fMRI allow for visualization of the hemodynamicactivity (during and after associative learning) in thecircuitry mediating learning with a good degree of spa-tial and temporal localization.

METHODS

Subjects

Four healthy young volunteers (two males, two fe-males, average age � 27) recruited from the North-western University community served as subjects. In-formed consent was obtained for participation in thestudy in keeping with guidelines approved by the in-stitutional review board. Subjects received neutral in-structions that would not reveal the nature of the as-sociation between the tone and airpuff during theexperiment. In addition, a movie, Milo and Otis, wasplayed to minimize boredom and prevent the dissipa-tion of arousal during the experimental session. Allsubjects received payment for their participation.

Behavioral Design and Delivery

The apparatus used for delivery of the stimulus wasa modified version of that used for eyeblink condition-ing in the rabbit (Akase et al., 1994; Thompson et al.,1994). The conditioning paradigm used to test themethodology of EMG detection was the delay 1250-ms

eyeblink conditioning paradigm outlined in Fig. 1. Thisparadigm has been shown to elicit robust learning inboth young and aging subjects (Knuttinen et al., 2001).The CS was a 1350-ms, 85-dB, 1000-Hz, 5-ms rise/falltone delivered binaurally over nonmagnetic acousti-cally shielded earphones (Avotec, Jensen Beach, FL).Auditory stimulus calibration occurred prior to testingeach subject using a constant tone and a Radio ShackRealistic sound level meter (No. 33–2050) with “C”weighting and slow response settings. The US was a4-psi 100-ms corneal puff of nitrogen gas delivered tothe right eye. A nitrogen tank was placed within thecontrol room and connected to a three-way solenoidvalve (Fluid Process Controls, Burr Ridge, IL, 24 V dc,No. 3823A2TVTF6). The airpuff stimulus was calibratedwith a digital pressure gauge (R&D Separations,Rancho Cordova, CA). The CS and US coterminated. Thetrigger signal was received from the computer runningLab View software. Stiff plastic tubing attached to thetank measuring 25 ft in length was passed through awaveguide into the magnet room. The delay betweenthe command pulse for the solenoid to open and theoccurrence of the puff at the eyelid was measured usingreflected light from a piece of tissue paper at the end ofthe delivery system. This delay was then entered intothe computer and the US onset time was adjusted tomaintain the proper timing of the tone and airpuff. Theend of the plastic tubing was fitted with a corrugateddrinking straw and pipet tip that was attached to thehead coil for precise alignment and delivery of theairpuff to the center of the right eye. The air pressureat the pipet tip was calibrated before each subject wastrained to ensure a 4-psi airpuff was being delivered.

Subjects received an initial block of 30 explicitlyunpaired random presentations of tone-alone and air-puff-alone trials to test for pseudoconditioning. Thesewere presented prior to 60 paired presentations of thetone and airpuff. The intertrial interval for all trialsranged from 16 to 20 s. A long intertrial interval wasused to allow the hemodynamic response to return tobaseline before the next trial. The initial pseudocondi-

FIG. 1. The electromyogram for eyeblink detection methodology was tested in the delay 1250-ms conditioning paradigm. This paradigmconsisted of a tone (CS), indicated by the hatched gray region, that overlapped and coterminated with the corneal airpuff unconditionedstimulus (US), indicated in black.

978 KNUTTINEN ET AL.

tioning trials served as a control baseline condition todetermine unconditioned response (UR) amplitudes,basal rates for eyeblinks to CS-alone presentations,and the hemodynamic response to sensory stimuli(tone and airpuff) alone. A smaller number of unpairedtrials were used due to the robust sensory responseevoked by the stimuli and to avoid latent inhibitioneffects, which can delay acquisition of learning (Siddleet al., 1987). A parametric behavioral study has dem-onstrated that relatively fast and robust learningcurves occur after the conditioning trials begin usingthis procedure (Carrillo et al., 1997). It was necessaryfor the pseudoconditioning trials to precede condition-ing since baseline measurements taken after the con-

ditioning would not have been at baseline level. Theywould have been affected by conditioning and wouldhave measured extinction processes.

Eyeblink Detection

The detection of the eyeblinks was accomplished byusing a device designed to monitor the electrocardio-gram (ECG) signal in the MRI environment (Physio-logic Monitor No. 3500, MR Equipment, Bayshore,NY). This device uses fiberoptic technology to reducethe noise induced during the scanning procedure. Ablock diagram of the setup is shown in Fig. 2.

FIG. 2. Block diagram showing the location of the electrodes and the electric/optical converters. The specifics of the signal conditioningare given. Note that the biopotential signal is primarily an optical signal while inside the magnet room. This reduces the potential fordistortion of the measured electromyographic signal.

979EYEBLINK CONDITIONING WITH fMRI

Eyeblinks were recorded via radiolucent ECG elec-trodes (Red Dot No. 2570, 3M, St. Paul, MN) placedaround the orbicularis oculi muscle of the right eye.Prior to electrode placement, the area was cleaned bythe experimenter with NuPrep gel (Grass InstrumentDivision of Astro-Med Inc., West Warwick, RI) to de-crease skin resistance. One electrode was placed 1 cmlateral to the outer canthus and a second was placed 1cm below the right eye. The ground electrode wasaligned at the center of the subject’s forehead. Therewas no overt motion of the electrodes during the eye-blink. The ECG monitoring device used a method sim-ilar to that described by Felbinger et al. (1996) to detectand conduct the signal out of the magnet room. Inshort, the electrical potentials were detected via theelectrodes and fed to the electro-optical converter,which was located just outside of the head coil. Theconverter was located as close as possible to the sub-ject’s head to minimize the length of the carbon-fiberwires (2 at 12.5 in. and one at 9.25 in.) leading to theelectrodes to reduce the amount of interference in-duced by the MRI scanner and far enough away toeliminate any susceptibility artifacts. The converterused a low-pass filter (23-Hz cutoff) and an amplifierwith a gain of 100 to improve the signal strength beforeconversion to an optical signal (see Fig. 2). The twobiopotentials, referenced to the ground electrode, weremultiplexed onto the same fiberoptic cable. The fiber-optic cable was passed through a waveguide in the RFshielding of the magnet room. Once outside of themagnet room, the optical signal was converted back toan electrical signal without any conditioning. The an-alog electrical signal (pin 11 referenced to pin 1 on thephysiologic monitor) was converted to a digital signalusing custom software written in Lab View (NationalInstruments, Austin, TX).

Lab View was used to coordinate the stimulus pre-sentation and processing of the behavioral response,which was sampled at 5 kHz. The postprocessing inLab View consisted of full wave rectification and inte-gration with a 10-ms time constant to generate apseudoanalog signal (see Fig. 3). The mean and stan-dard deviation (SD) of the 500-ms baseline waveformwere calculated. A conditioned response was one thatexceeded the mean baseline value by 4 SD for a mini-mum of 10 ms prior to US onset.

Noise Assessment of the ECG Device

It is imperative to minimize the amount of noiseinduced in the fMRI signal time series data. Therefore,before implementing the study described, a noise as-sessment of the ECG device was completed. This con-sisted of a functional run with and without the devicepresent. The experimental runs used the same imagingparameters as the fMRI acquisition except that thenumber of volumes was limited to 128 after the signal

had reached steady state. A phantom was used to as-sess the overall noise and a human volunteer was usedto determine if the noise was detectable above thatinduced by physiological processes in the brain. Theelectrodes were placed on the phantom in a fashionthat simulated the method described above. In thevolunteer study, the electrodes were applied as de-scribed. During the experiment no behavioral stimuliwere given to detect the noise associated with the de-vice. The subject was asked to lie quietly during theprocedure.

The time course data were analyzed to determine ifthe variance changed with the presence of the ECGdevice. It is possible to use the �2 statistic to test thehypotheses about the variance. However, if one as-sumes that the data are normal and a sufficiently largesample is used (n � 40), it is possible to use a Z statisticto test the differences in variance (Devore, 1982). Thisstatistic can be written as

Z �S � �0

�0 /�2n,

where S and �0 are the standard deviations of the twodifferent measures, and n is the number of samples. Tocontrol for Type I errors, an � � 0.01 was used that

FIG. 3. Example of a conditioned response recorded inside themagnet. (A) Rectified and integrated (� � 10 ms) electromyographicactivity for a single trial. The amplitude is shown in relative arbi-trary integrated units. (B) The “raw” electromyographic signal usedto generate the signal in (A). The amplitude is shown as increasesand decreases in relative voltage. (C) Relative timing of the toneconditioning stimulus and airpuff unconditioned stimulus.


corresponds to a Z value of 2.33. The number of sam-ples in the time course (n) was 128 and the variancewas derived from the region of interest (ROI) in eachdata set. The variance for the data set without thedevice present was the reference value, �0. A �Z� � 2.33would indicate that the device had a significant effecton the time course data. The ROI was located at thelevel and in the same hemisphere (right) as the electro-optical converter, since the maximum noise is assumedto occur nearest to the converter. In the volunteer, theROI included both gray and white matter, while ex-cluding the ventricles. Similar analysis was completedon the phantom data. A power spectral analysis wasconducted on the same data similar to that describedby Zarahn et al. (1997) to observe changes in the noisespectrum due to the presence of the ECG device. Theresults of the power spectral analysis were reviewedqualitatively.

Scanning Procedures

Imaging was conducted at Northwestern UniversityMedical School on a Siemens Vision 1.5-T scannerequipped with whole-body EPI gradients and a trans-mit/receive quadrature head coil. Head movement wasminimized with the use of a vacuum pillow (Vac-Fix,Toledo, OH) and restraint calipers built into the headcoil. These devices help to keep movement minimizedto � 1 mm total excursion during the functional run(20 min) and the intervolume deviation to submillime-ter deviations (Parrish et al., 1998). fMRI data werecollected using an EPI sequence with parameters ofTR � 2000 ms, TE � 40 ms, FOV � 240 mm, matrix of64 � 64, and 24 axial slices of 6-mm thickness. Imagesof whole brain were acquired with slices oriented par-allel to the anterior commissure–posterior commissure(AC/PC) plane. Imaging consisted of two functionalruns: a pseudoconditioning run, followed by a condi-tioning run. The pseudoconditioning run included atotal of 60 unpaired trials. The subsequent condition-ing run included a total of 60 paired trials, which lastedapproximately 20 min. High-resolution 3D anatomicalMRI data were collected at the end of the functionalseries.

Behavioral Data Analysis

The criteria for an eyeblink response to be classifiedas a conditioned response (CR) required the integratedsignal to: (1) be greater than or equal to 4 SD above thebaseline for a minimum duration of 10 ms; (2) occurmore than 100 ms after tone onset (to correct for vol-untary responses); and (3) remain above baselinethroughout the CS–US interval before blending intothe unconditioned response (UR). Eyeblinks that re-turned to baseline before US onset were defined asalpha responses. In general, these alpha responsestended to be of short latency and duration. The same

response criteria were applied to the pseudocondition-ing tone-alone trials to that portion of the trial prior towhen the US would have occurred. Following eachtraining session, the computer generated a graphicaldepiction and a tabulated summary of the eyeblinkresponses for each trial. The overall percentage of tri-als with CRs was determined for each subject. The datawere also used to determine differences in the amountof conditioning (percentage CRs) between the pseudo-conditioning and conditioning trials.

Functional Data Analysis

Image data were transferred to an HP workstationfor analysis using SPM97 software (Wellcome Depart-ment of Cognitive Neurology, London, UK). Time seriesdata were acquisition timing corrected and motion cor-rected. For the standard analysis, a time series of eachvoxel was correlated with a reference waveform andtransformed into a Z score map (Friston et al., 1994,1995). A model of the expected magnetic resonancesignal response was created by convolving the stimuluswith a canonical hemodynamic response function (Fris-ton et al., 1995). In this experiment, the modeled stim-ulus period included both the tone and airpuff, whichlasted less than 2 s. The general linear model was usedto determine the statistical significance on a voxel-by-voxel basis (Friston et al., 1994). In addition, the re-sponse from the pseudoconditioning phase was sub-tracted from that of the conditioning phase to showonly differences in areas underlying associative learn-ing. Activation maps of the difference data were madeusing an uncorrected threshold of P � 0.001 and anextent threshold of 3 voxels.

RESULTS

For all subjects, the EMG signal was successfullyrecorded during fMRI scanning using the 3M elec-trodes, NuPrep gel, and physiologic monitor (see Fig.3). The EPI sequence added minimal artifacts to theEMG signal and the filtering reduced them even fur-ther as demonstrated in Fig. 3. Figure 4 shows theartifact-free signals obtained from a learner and a non-learner, while their eyeblink responses were beingmonitored in the magnet. Furthermore, the EMG sig-nal clearly detected differences between trials in whichthe subject showed a conditioned response (eyeblink) tothe tone and those trials in which the subject did notshow such responses. In all four of our subjects, theEMG signals were suitable for detecting eyeblink mea-surements, which correlated with the behavioral par-adigm being used, i.e., EMG detection of eyeblinks(URs) was present on all trials and in all subjects whenan airpuff was given. The EMG electrodes, carbon fiberleads, and electro-optical converter located outside ofthe head coil (see Fig. 2) did not produce any visually


detectable susceptibility artifacts in the images ac-quired. However, it was our experience that there wasnoticeable signal loss in the brain areas adjacent tothe converter, if the converter was placed within thehead coil.

The time series of signal intensity from the ROI withthe mean removed and the power spectrum generatedfrom the phantom data are shown in Fig. 5. Using thestatistical analysis described in the Methods section, aZ score was calculated to determine if the variance ofthe signal intensity with and without the ECG devicepresent was the same. The time course of the phantomdata with the mean removed is shown in Fig. 5A. The

Z scores were 1.2 and 0.24 for the phantom data andthe human data, respectively, indicating there was nostatistical difference. Therefore, the ECG device didnot add any detectable levels of noise to the images. Alower Z score indicated that the time series were fromthe same population. The Z score for the human datawas much lower than that for the phantom data, whichindicated that the time series with and without theECG device were nearly identical. It is hypothesizedthat this is due to the presence of physiologic noise thattends to dominate the fMRI time course. Furthermore,the power spectrum from the phantom data demon-strates the relative power of the noise at the frequen-cies properly sampled by the TR of 2 s (Fig. 5B). Theseresults indicate that the presence of the ECG devicedoes not alter the noise spectrum. The same was alsotrue for the human data.

All subjects reported that they were able to hear thetone through the nonmagnetic headphones and wereable to feel the airpuff, which remained aimed at theright eye when observed at the end of the imagingsession. The motion correction results demonstratedthat subjects moved less than 1 mm during an entireimaging run (20 min). In addition, it did not appearthat there was any stimulus-correlated head motion.

Two subjects were able to condition, i.e., learn toassociate the CS with the subsequent US, at levels of60 and 80% CRs by the end of the conditioning run. Theconditioning run was later divided into three blocks of20 trials to look at acquisition of learning, i.e., anincrease in CRs across a given session. The percentageof CRs was greater during the second and third blocksof conditioning trials (50 and 70%) in these learnersthan after the first block of conditioning (20%), indicat-ing that substantial associative learning did occur astrials progressed (Fig. 6). Two subjects were not able tocondition: one had progressively smaller URs and re-ported having fallen asleep during the experimentalsession (�10% CRs), and the other was unable to reacha level beyond 20% CRs (either inside or outside of themagnet). Therefore, the lower conditioning level can-not be attributable to factors associated with the scan-ning environment. All subjects reported being comfort-able throughout the training session.

In addition to showing that the functional imagesacquired were not contaminated by artifact from theEMG methodology, an analysis was done in all subjectsto investigate the brain regions that became activatedduring conditioning. Activation maps for conditioningwere generated by using cognitive subtraction of thepseudoconditioning (both CS and US) baseline fromthe conditioning trials to remove baseline and nonas-sociative sensory activation.

Figure 7 depicts the brain areas in the learners thatwere significantly more active during the paired tone–airpuff trials than during the pseudoconditioning un-paired trials. As shown, areas of significant activation

FIG. 4. Example of filtered, rectified, and integrated electromyo-graphic tracings in the delay 1250-ms task. The electromyographicresponse was taken from a single trial during the conditioning run,while magnetic resonance images were being acquired. Tone onsetoccurred at 500 ms, and subsequently overlapped and coterminatedwith the US. The gray-shaded area represents the area of the USstimulus presentation (1750–1850 ms). Each trial was monitored fora duration of 2500 ms. Note the relatively low level of baselineactivity in both the top and bottom panels, indicating minimal dis-turbance of the electromyographic signal by the magnetic resonancepulse sequences. (A) Typical conditioned response observed in twosubjects who were able to acquire the delay 1250-ms task. (B) Typicalresponse elicited by a subject who was unable to acquire the delay1250-ms task even after receiving 60 conditioning trials. Note thelack of response to the tone CS. The subject did, however, show arobust UR.


FIG. 5. Noise analysis of the time series data from a phantom with and without the ECG device. (A) Time course of the magneticresonance signal intensity with and without the device present is shown. Using the analysis described in the text, the Z score was 1.2, whichis less than the critical Z of 2.33 needed to determine if the device caused a significant amount of noise. (B) Power spectrum of the noise. Onlyfrequencies (0–0.25 Hz) that are properly sampled by the TR � 2 s are shown. Note that the two curves are nearly identical, indicating thatthere is no change in the noise spectrum or any increase in the noise due to the presence of the ECG device.


included bilateral cerebellum, auditory cortex, basalganglia, posterior cingulate, and bilateral frontal cor-tex. The two nonlearners showed activations in audi-tory and visual cortex during the pseudoconditioningcontrol condition, but neither of these two subjectsshowed any areas of activation in the conditioning mi-nus pseudoconditioning.

DISCUSSION

Electromyography as a recording system allows forexcellent detection of eyeblink responses during echoplanar functional imaging that is equivalent to eye-blink detection in conditioning experiments outside ofthe magnet. This experimental setup confirms the find-ings of Felbinger et al. (1996) that showed that thereare no additional artifacts on the EMG signal imposedby echo planar sequences. These artifacts are mini-mized by the short carbon-fiber wires, eliminating anyloops, and the placement of the electro-optical con-verter just outside the head coil. Figure 5 demonstratesthat there is no added noise generated by the physio-logic monitor. Further analysis of the noise data dem-onstrated that physiologic noise from the subject dom-inated the noise present in the data. The behavioralsetup described in this article, therefore, allows foradequate and safe monitoring of behavioral eyeblinkresponses that occur during classical eyeblink condi-tioning and serves as a helpful tool that may be used inother behavioral experiments that require the monitor-ing of eyeblinks.

Unilateral measurements of eyeblink responses isadequate in most eyeblink conditioning experiments;however, two physiologic monitors could be used tomonitor bilateral eyeblinks, if desired. The use of fiber-optic for the transmission of signal outside the magnetbore also decreases the potential of electromagneticartifacts. The electrodes were placed at least 1 cmaway from the outer canthus of the eye, to avoid thepotential risk of heating hazards (which did not occur

in our experiments). This placement is adjacent to theorbicularis oculi muscle and has been successfully usedin behavioral experiments conducted outside of themagnetic resonance environment (Knuttinen et al.,2001). Given the magnitude of the EMG responsesobtained in the subjects in this experiment, it is clearthat this electrode placement was sufficient for therobust detection of eyeblink responses.

Our results from this methodological study alsoshowed that classical eyeblink conditioning serves as auseful behavioral paradigm to be used in conjunctionwith event-related fMRI, which allows for the targetingof specific memory processes. Our findings of the learn-ing-specific activations are generally consistent withresults reported previously in positron emission tomog-raphy (PET) studies (Blaxton et al., 1996; Logan andGrafton, 1995; Molchan et al., 1994; Schreurs et al.,1997). For example, there is substantial precedent forthe observation of changes observed in cerebellum.There is a significant amount of animal literature thatdocuments the importance of the role of the cerebellumas a convergence point for the CS and the US in eye-blink conditioning (Lavond et al., 1993; Schreurs et al.,1995) and in the processing of the temporal relation-ship between stimuli critical for associative learning(Topka et al., 1993; Perret et al., 1993). Both subjectswho demonstrated an increase in conditioned re-sponses during the training session showed increasedactivation in the ipsilateral cerebellum in the pairedstate compared with the unpaired control condition.This indicates that the changes were specifically re-lated to learning, rather than to the nonspecific aspectsof the experimental protocol. These data are consistentwith the previous literature of human lesion studiesthat have demonstrated impaired conditioning in pa-tients with cerebellar damage (Solomon et al., 1989)and with rabbit studies that show an essential role forthe ipsilateral cerebellum in eyeblink conditioning(McCormick and Thompson, 1984; Krupa et al., 1993).One of our subjects showed substantial blood flow in-creases in bilateral cerebellum that are also consistentwith studies in rabbits that have shown that bilateralbehavioral responses occur when the airpuff is deliv-ered to one eye (Disterhoft et al., 1977). Furthermore,although unilateral lesions of cerebellar cortex impairconditioning, considerable transfer of training does oc-cur when rabbits are subsequently trained on the non-lesioned and previously naı̈ve side (McCormick et al.,1982; Yeo et al., 1985; Lavond et al., 1994).

Patterns of learning-specific regional activation inthe two subjects who learned were also observed in CSsensory auditory cortex, basal ganglia, and posteriorcingulate. Conditioning-specific activation of the audi-tory cortex contralateral to the side of the airpuff haspreviously been documented in other functional ana-tomical studies of associative learning in humans(Bahro et al., 1999; Molchan et al., 1994; Schreurs et

FIG. 6. Performance of learners versus nonlearners in the delay1250-ms eyeblink paradigm. The conditioning run was subdividedinto three blocks of 20 trials (for a total of 60 conditioning trials) toobserve the rates of learning in the single run. Learners were able toreach a level of 70% CRs by the end of conditioning. The nonlearners,however, were unable to reach a level past 20% CRs.


al., 1997). Lesions of the basal ganglia have been re-ported to impair the eyeblink response in rabbits (Kaoand Powell, 1988), and basal ganglia activation hasbeen observed in PET studies of human eyeblink con-ditioning (Blaxton et al., 1996). Cohen and Eichen-baum (1993) suggest the basal ganglia, in associationwith the cerebellum, is an important contributor to theformation of associations between related events, suchas the CS and US. One important area we found to be

substantially activated was the posterior cingulate.The role of the cingulate region is heterogeneous andincludes processes of attention, learning, and memory(Mesulam, 2000; Gabriel, 1990). Bahro et al. (1999)showed an increase in regional cerebral blood flow inthe area of posterior cingulate with PET, which repli-cated their previous PET study findings (Molchan etal., 1994). Our findings support their work and thehypothesis that the cingulate region is a component of

FIG. 7. The activation maps for conditioning were generated by subtracting the pseudoconditioning (CS and US alone) baseline from theconditioning trials (CS and US paired) to remove sensory-related activation. The pseudoconditioning maps were constructed from the 30unpaired random presentations of tone alone and airpuff alone trials. Areas of increased activation relative to the pseudoconditioningbaseline include: areas of cerebellum and caudate, cingulate, temporal, and frontal cortex. The nonlearners showed activity in thepseudoconditioning run, but both of these subjects showed no areas of activation in the subtraction maps.


the paralimbic system. In particular, the posterior cin-gulate region receives the majority of hippocampal pro-jections (Mesulam, 2000), thereby making it an impor-tant contributor in learning and memory tasks. Inaddition, animal studies have also documented the cin-gulate cortex to be reciprocally connected with all sen-sory cortices, making it an ideal area for the integra-tion of stimulus associations (Gabriel, 1990; Papernaand Malach, 1991). The results we obtained in ourconditioned subjects indicate that the methods out-lined here offer a new and exciting approach withwhich to further examine the processes that may un-derlie associative learning in humans.

In summary, the goal of this study was to adapt andoptimize a methodological approach for human classi-cal eyeblink conditioning to be used in conjunction withevent-related fMRI. This study indicates that eyeblinkconditioning can be done using fMRI, despite the com-plex nature of the behavioral setup within the magnet.In addition, this methodology has recently been used torecord behavioral responses without causing artifactsusing a 3-T scanner (Preston et al., 2000).

Although several PET studies have examined differ-ences in blood flow in the human brain following con-ditioning, those studies are limited in their spatialresolution and repeatability (due to radioactivity) ascompared with fMRI. The approach we describe hereoffers the opportunity to further extend the humaneyeblink conditioning studies through the use of fMRIto better visualize those brain regions that show hemo-dynamic alteration during associative learning.

Finally, the analysis described in this study may befurther modified to investigate other components of thelearned response. One example would be to look atspecific behavioral performance in individual learningtrials, i.e., identify trials on which subjects produced aCR and contrast them with trials without CRs. Inaddition, this method is a valuable tool to look at mem-ory consolidation since the same subjects can return forimaging after a period of time. Theoretically, brainregions activated during recall of the consolidatedmemories can be visualized.

ACKNOWLEDGMENTS

The authors thank S. J. Weiss and M. M. Oh for their assistancewith this project.

REFERENCES

Akase, E., Thompson, L. T., and Disterhoft, J. F. 1994. A system forquantitative analysis of associative learning: Real-time softwarefor MS-DOS microcomputers. J. Neurosci. Methods 54: 119–130.

Bahro, M., Molchan, S. E., Sunderland, T., Herscovitch, P., andSchreurs, B. G. 1999. The effects of scopolamine on changes inregional cerebral blood flow during classical conditioning of thehuman eyeblink response. Neuropsychobiology 39: 187–195.

Blaxton, T. A., Zeffiro, T., Gabrieli, J., Bookheimer, S. Y., Carrillo,M., Theodore, W., and Disterhoft, J. F. 1996. Changes in cerebral

blood flow during eyeblink conditioning in humans. J. Neurosci.16: 4032–4040.

Carrillo, M. C., Thompson, L. T., Gabrieli, J. D. E., and Disterhoft,J. F. 1997. Variation of the intertrial interval in human classicalconditioning. Psychobiology 25: 152–157.

Clark, R. E., and Squire, L. R. 1998. Classical conditioning and brainsystems: The role of awareness. Science 280: 77–81.

Cohen, M. S., and Bookheimer, S. Y. 1994. Localization of brainfunction using magnetic resonance imaging. Trends Neurosci. 17:268–277.

Cohen, N. J., and Eichenbaum, H. 1993. Memory, Amnesia and theHippocampal System. MIT Press, Cambridge, MA.

Daum, I., and Schugens, M. M. 1996. On the cerebellum and classicalconditioning: Current directions. Psychol. Sci. 5: 58–61.

Daum, I., Schugens, M. M., Ackermann, H., Lutzenberger, W.,Dichgans, J., and Birbaumer, N. 1993. Classical conditioning aftercerebellar lesions in humans. Behav. Neurosci. 107: 748–756.

Demb, J. B., Desmond, J. E., Wagner, A. D., Vaidya, C. J., Glover,G. H., and Gabrieli, J. D. E. 1995. Semantic encoding and retrievalin the left inferior prefrontal cortex: A functional MRI study of taskdifficulty and process specificity. J. Neurosci. 15: 5870–5878.

Desmond, J. E., Gabrieli, J. D. E., Wagner, A. D., Ginier, B. L., andGlover, G. H. 1997. Lobular patterns of cerebellar activation inverbal working memory and finger tapping tasks as revealed byfunctional MRI. J. Neurosci. 17: 9675–9685.

Devore, J. L. 1982. Probability & Statistics for Engineering and theSciences. Brooks/Cole, Monterey, CA.

Disterhoft, J. F., Kwan, H. H., and Lo, W. D. 1977. Nictitatingmembrane conditioning to tone in the immobilized albino rabbit.Brain Res. 137: 127–143.

Felbinger, J., Muri, R., Ozdoba, C., Schroth, G., Hess, C. W., andBoesch, C. 1996. Recordings of eye movements for stimulus controlduring fMRI by means of EOG methods. Magn. Reson. Med. 36:410–414.

Friston, K. J., Jezzard, P., and Turner, R. 1994. Analysis of func-tional MRI time-series. Hum. Brain Mapp. 1: 153–171.

Friston, K. J., Frith, C. D., Turner, R., and Frackowiak, R. S. 1995.Characterizing evoked hemodynamics with fMRI. NeuroImage 2:157–165.

Gabriel, M. 1990. Functions of anterior and posterior cingulate cor-tex during avoidance learning in rabbits. Prog. Brain Res. 85:467–483.

Gabrieli, J. D. E., McGlinchey-Berroth, R., Carrillo, M. C., Gluck,M. A., Cermak, L. S., and Disterhoft, J. F. 1995. Intact delay-eyeblink classical conditioning in amnesics. Behav. Neurosci. 109:819–827.

Gabrieli, J. D. E., Desmond, J. E., Demb, J. B., Wagner, A. D., Stone,M. V., Vaidya, C. J., and Glover, G. H. 1996. Functional magneticresonance imaging of semantic memory processes in the frontallobes. Psychol. Sci. 7: 278–283.

Gabrieli, J. D. E., Brewer, J. B., Desmond, J. E., and Glover, G. H.1997. Separate neural bases of two fundamental memory pro-cesses in the human medial temporal lobe. Science 276: 264–266.

Kao, K. T., and Powell, D. A. 1988. Lesions of the substantia nigraretard Pavlovian eye-blink but not hear trate conditioning in therabbit. Behav. Neurosci. 102: 515–525.

Knuttinen, M.-G., Weiss, C., Parrish, T. B., LaBar, K. S., Gitelman,D. R., Power, J. M., Mesulam, M.-M., and Disterhoft, J. F. 2000.Event-related fMRI of delay eyeblink conditioning. NeuroImage11: S416.

Knuttinen, M.-G., Power, J. M., Preston, A. R., and Disterhoft, J. F.2001. Awareness in classical differential eyeblink conditioning inyoung and aging humans. Behav. Neurosci. 115: 747–757.


Krupa, D. J., Thompson, J. K., and Thompson, R. F. 1993. Localiza-tion of a memory trace in the mammalian brain. Science 260:989–991.

Kwong, K. K. 1995. Functional magnetic resonance imaging withechoplanar imaging. Magn. Reson. Q. 11: 1–20.

Lavond, D. G., Kim, J. J., and Thompson, R. F. 1993. Mammalianbrain substrates of aversive classical conditioning. Annu. Rev.Psychol. 44: 317–342.

Lavond, D. G., Kanzawa, S. A., Ivkovich, D., and Clark, R. E. 1994.Transfer of learning but not memory after unilateral cerebellarlesions in rabbits. Behav. Neurosci. 108: 284–293.

LeBihan, D., and Karni, A. 1995. Applications of magnetic resonanceimaging to the study of human brain function. Curr. Opin. Neuro-sci. 5: 231–237.

Logan, C. G., and Grafton, S. T. 1995. Functional anatomy of humaneyeblink conditioning determined with regional cerebral glucosemetabolism and positron-emission tomography. Proc. Natl. Acad.Sci. USA 92: 7500–7504.

McCormick, D. A., Guyer, P. E., and Thompson, R. F. 1982. Superiorcerebellar peduncle lesions selectively abolish the ipsilateral clas-sically conditioned nictitating membrane/eyelid response of therabbit. Brain Res. 244: 347–350.

McCormick, D. A., and Thompson, R. F. 1984. Neuronal responses ofthe rabbit cerebellum during acquisition and performance of aclassically conditioned nictitating membrane/eyelid response.J. Neurosci. 4: 2811–2822.

McGlinchey-Berroth, R., Carrillo, M. C., Gabrieli, J. D., Brawn,C. M., and Disterhoft, J. F. 1997. Impaired trace eyeblink condi-tioning in bilateral, medial-temporal lobe amnesia. Behav. Neuro-sci. 111: 873–882.

Mesulam, M.-M. 2000. Principles of Behavioral and Cognitive Neu-rology. Oxford Univ. Press, New York.

Molchan, S. E., Sunderland, T., McIntosh, A. R., Herscovitch, P., andSchreurs, B. G. 1994. A functional anatomical study of associativelearning in humans. Proc. Natl. Acad. Sci. USA 91: 8122–8126.

Paperna, T., and Malach, R. 1991. Patterns of sensory intermodalityrelationships in the cerebral cortex of the rat. J. Comp. Neurol.308: 432–456.

Parrish, T. B., Gitelman, D. R., Kim, Y.-H., LaBar, K. S., Hallam, D.,and Mesulam, M.-M. 1998. Clinical fMRI: Is patient motion reallyan issue? NeuroImage 7: S560.

Perrett, S. P., Ruiz, B. P., and Mauk, M. D. 1993. Cerebellar cortexlesions disrupt learning-dependent timing of conditioned eyelidresponses. J. Neurosci. 13: 1709–1718.

Poldrack, R. A., Desmond, J. E., Glover, G. H., and Gabrieli, J. D. E.1998. The neural basis of visual skill learning: An fMRI study ofmirror reading. Cereb. Cortex 8: 1–10.

Preston, A. R., Knuttinen, M.-G., Christoff, K., Glover, G. H., Gab-rieli, J. D. E., and Disterhoft, J. F. 2000. The neural basis ofclassical eyeblink conditioning: An event-related fMRI study. Soc.Neurosci. Abstr. 26: 709.

Ramnani, N., Toni, I., Josephs, O., Ashburner, J., and Passingham,R. E. 2000. Learning- and expectation-related changes in the hu-man brain during motor learning. J. Neurophysiol. 84: 3026–3035.

Schreurs, B. G., Oh, M. M., Hirashima, C., and Alkon, D. L. 1995.Conditioning-specific modification of the rabbit’s unconditionednictitating membrane response. Behav. Neurosci. 109: 24–33.

Schreurs, B. G., McIntosh, A. R., Bahro, M., Herscovitch, P., Sun-derland, T., and Molchan, S. E. 1997. Lateralization and behav-ioral correlation of changes in regional cerebral blood flow withclassical conditioning of the human eyeblink response. J. Neuro-physiol. 77: 2153–2163.

Siddle, D. A. T., and Remington, B. 1987. Latent inhibition andhuman Pavlovian conditioning. In Cognitive Processes and Pavlov-ian Conditioning in Humans (G. Davey, Ed.), pp. 115–146. Wiley,New York.

Solomon, P. R., Stowe, G. T., and Pendlebury, W. W. 1989. Disruptedeyelid conditioning in a patient with damage to cerebellar affer-ents. Behav. Neurosci. 103: 898–902.

Thompson, L. T., Moyer, J. R., Jr., Akase, E., and Disterhoft, J. F.1994. A system for quantitative analysis of associative learning. 1.Hardware interfaces with cross-species applications. J. Neurosci.Methods 54: 109–117.

Thompson, R. F., Berger, T. W., Cegavske, C. F., Patterson, M. M.,Roemer, R. A., Teyler, T. J., and Young, R. A. 1976. A search for theengram. Am. Psychol. 31: 209–227.

Topka, H., Valls-Sole, J., Massaquoi, S. G., and Hallett, M. 1993.Deficit in classical conditioning in patients with cerebellar degen-eration. Brain 116: 961–969.

Wagner, A. D., Desmond, J. E., Demb, J. B., Glover, G. H., andGabrieli, J. D. E. 1997. Semantic repetition priming for verbal andpictorial knowledge: A functional MRI study of left inferior pre-frontal cortex. J. Cogn. Neurosci. 9: 714–726.

Woodruff-Pak, D. S. 1988. Aging and classical conditioning: Parallelstudies in rabbits and humans. Neurobiol. Aging 9: 511–522.

Woodruff-Pak, D. S. 1993. Eyeblink classical conditioning in H.M.:Delay and trace paradigms. Behav. Neurosci. 107: 911–925.

Yeo, C. H., Hardiman, M. J., and Glickstein, M. G. 1985. Classicalconditioning of the nictitating membrane response of the rabbit. II.Lesions of the cerebellar cortex. Exp. Brain Res. 60: 99–113.

Zarahn, E., Aguirre, G. K., and D’Esposito, M. 1997. Empirical anal-yses of BOLD fMRI statistics. NeuroImage 5: 179–197.


Electromyography as a Recording System for Eyeblink Conditioning with Functional Magnetic Resonance Imaging

Documents